BACKGROUND OF THE INVENTION
1. Field of the Invention:
[0001] This invention relates to a microprocessor suitable for graphics process, and more
particularly to graphics processing apparatus and method using the microprocessor
and suitable for transferring graphics data between memories on separate buses.
2. Description of the Related Art:
[0002] A conventional microprocessor system having two sets of buses, each set for address,
data and control signals, is exemplified by Harvard Architecture. This conventional
system is designed so as to avoid any contention between an instruction fetch and
a data access by separating an instruction bus and a data bus.
[0003] Also, LSIs for graphics process are exemplified by a bit map control processor (BMCP)
discussed in
Toshiba Review 43th volume, 12th issue (1988), pages 932-935. This BMCP has an 8-bit data bus, as a system bus to which a CPU and a system memory
are connected, and a 64-bit local memory data bus and a 24-bit address bus independently
of the data bus; the address bus and local memory data bus access 8-plane local memories
(image memories). Further, with an address latched, the BMCP can access the system
memory.
[0004] In the Harvard Architecture, the instruction bus and the data bus are dedicated;
the Harvard Architecture is totally silent about the concept of using two sets of
buses in data access, such as in graphics transfer between a system memory and a frame
memory which transfer is performed most frequently among various procedures of the
graphics processing.
[0005] The BMCP may use two sets of buses in data access, but has only a single address
bus; it is unclear from Toshiba Review that two memories can be simultaneously accessed.
[0006] Further, in executing a so-called read/modify/write instruction frequently used for
graphics processing in which data in a memory address is read and processed and is
then written back to the same address, conventional general microprocessors have the
following problems.
[0007] First of all, in CISC (Complex Instruction Set Computer) type general processors,
although a single instruction is capable of describing a read/modify/write operation
with memory addresses designated for a source operand and a destination operand of
the operation instruction, the length of instruction is necessarily long. Also, it
is not clear if it is possible to make the execution without interposing a wait cycle
between a read cycle and a write cycle.
[0008] In some of exclusive processors for graphics process, though a read/modify/write
instruction can be executed in two successive memory cycles, its modifying function
is limited.
[0009] In RISC (Reduced Instruction Set Computer) type processors, instructions are basically
of a fixed length, and operands accessible to a memory are allowed to be designated
only for a load instruction and a store instruction. Thus, execution of a read/modify/write
operation requires three instructions, i.e. load, operation and store instructions,
failing to execute the load and store instructions in two successive memory cycles.
SUMMARY OF THE INVENTION
[0010] It is therefore an object of this invention to enable a high-speed transfer of graphics
data between two memories, as accessed simultaneously to the memories on separate
buses, in a graphics processing apparatus for processing a large quantity of graphics
data.
[0011] Another object of the invention is to enable the
[0012] EP A-0 279 229 discloses a graphic display system. It comprises a display adaptor circuit connected
to a system processor by a bus. Further, the adaptor circuit is connected to a RGB
monitor by an output bus. The display adaptor circuit includes two memories connected
to a digital signal processor used for circuit resource management.
[0013] The paper "Digital signal processing IC helps to shed new light on image processing
applications" by Casey, P, and others, in "
Electronic Design", Penton publishing, Cleveland, Ohio, United States, March 20, 1986,
pages 135 to 140, discloses a chip having independent multiplier-accumulator, ALU and barrel shift
sections. These components receive input data from various bases and output their
results to a result bus. execution of a read/modify/write operation very popular in
graphics processing, without interposing a null cycle between a read cycle and a write
cycle, in a processor of RISC-type.
[0014] A graphics processing apparatus according to this invention is defined by claims
1-3.
[0015] With the arrangement of this invention, the two-port microprocessor performs the
reading, decoding and executing of an instruction and the storing of data, preferably
with pipeline processing. In the case of memory access instruction, its process is
assigned to a unit dedicated for memory access processing, whereupon the two-port
microprocessor executes the next instruction. If the process has already assigned
to the port, the memory access processing unit causes the instruction executing unit
to wait executing the next instruction, until the port becomes available for use,
by managing the operation status of the two ports. Therefore, even while one port
is in operation, the memory access via the other portion can be accepted.
[0016] Partly since two sets of buses of the two-port microprocessor are usable for instruction
fetch and data access, and partly since two simultaneous memory accesses are possible,
the memory accessing efficiency is improved. In the case of a graphics processing
apparatus in particular, it is possible to transfer the graphics data between the
system memory and the frame memory at high speed.
[0017] Further, with the read and write instructions being located in the program in the
memory, and with the operation process instruction being located in a sub-instruction
buffer independent of the ordinary instruction buffer, an operation instruction is
fetched from the above buffer at a timing at which the read data are ready to use,
so that the operation of the read data is in time for the write cycle, enabling a
read/modify/write process of the two successive memory cycles.
[0018] Parameters for graphics drawing, rectangular field transfer, with a logic operation
or the like are not given directly from the CPU to the two-port micro-processor, but
are successively written in the memory by the CPU, and at the same time, its leading
address is placed in the work list in the memory, so that subsequent graphics processing
such as graphics drawing and transfer can all be assigned to the two-port microprocessor.
Namely, as the graphics processing is started by the CPU, the two-port microprocessor
executes the designated processes in the work list, successively with reference to
the processing program and parameters. In the CPU, if a work list is prepared beforehand
for a plurality of graphics processes, it is unnecessary to give to the two-port microprocessor
parameters for each and every graphics process, thus minimizing the burden on the
CPU.
[0019] The above and other advantages, features and additional objects of this invention
will be manifest to those versed in the art upon making reference to the following
detailed description and the accompanying drawings in which several preferred embodiments
incorporating the principles of this invention are shown by way of illustrative example.
BRIEF DESCRIPTION OF THE DRAWINGS
[0020]
FIG. 1 is a block diagram of a graphics processing apparatus using a two-port microprocessor
according to this invention;
FIG. 2 is a memory map of the apparatus of FIG. 1;
FIG. 3 is a diagram showing internal registers of the processor of FIG. 1;
FIG. 4 is a memory map of commands and programs of the processor;
FIG. 5 is a flowchart showing the mode of operation of the processor;
FIG. 6 is a block diagram showing the interior of the processor;
FIG. 7 is a block diagram of a port number designating circuit in a decoder 1102 in
FIG. 6;
FIG. 8 is a table showing the mode of operation of the circuit of FIG. 7;
FIG. 9 is a block diagram of a register forced access executing circuit in the decoder
1102 of FIG. 6;
FIG. 10 is a block diagram showing the interior of a memory managing unit 1104 in
FIG. 6;
FIGS. 11 through 13 are memory access timing diagrams of the processor of FIG. 6;
FIGS. 14 through 16 are diagrams showing the manner in which a read/modify/write instruction
is executed;
FIG. 17 is a block diagram showing the concept of this invention using a modified
buffer;
FIG. 18 is a diagram showing the manner in which graphics data are transferred from
a system memory to a frame memory; and ,
FIG. 19 is a detailed block diagram showing an operation unit in a two-port microprocessor.
DETAILED DESCRIPTION
[0021] The principles of this invention are particularly useful when embodied in a graphics
processing apparatus (hereinafter called "apparatus") such as shown in FIG. 1. The
apparatus of this embodiment is suitable for use in processing a large quantity of
graphics data in a personal computer (PC), a printer, etc.
[0022] In FIG. 1, reference numeral 8 designates a central processing unit (CPU) for controlling
the entire apparatus.
[0023] 1 designates a two-port microprocessor for instructing a graphics drawing or a graphics
display on a CRT 6. The microprocessor 1 includes therein a processor part 101 for
executing a microinstruction, and two ports 102, 103.
[0024] 3 designates a local memory for storing a micro-instruction and data for graphics
drawing.
[0025] 4 designates a frame memory for storing graphics data.
[0026] 5 designates a shift circuit for outputting display data of a plurality of picture
elements or pixels read as the microprocessor 1 makes a graphics display access to
the frame memory 4, to the CRT 6 one pixel by one pixel.
[0027] 7 designates a clock generator for generating a clock signal for activating the two-port
microprocessor 1, the local memory 3, the frame memory 4 and the shift circuit 5.
[0028] 10 designates a system memory for storing microinstructions and data to be executed
by the CPU 8 and also for storing graphics drawing commands and data to be processed
by the microprocessor 1.
[0029] 11 designates a bus arbitrator for requesting a bus for the CPU 8 upon receipt of
a request signal, for use of the system bus 9, which signal is outputted by the microprocessor
1 before accessing to the system memory 10. Then, when a permission for use of the
bus has been given, the bus arbitrator 11 gives this information to the microprocessor
1.
[0030] 12 designates a decoder for decoding an address, outputted by the CPU 8, to output
to the microprocessor 1 an interrupt request and a type of interrupt processing when
the CPU 8 accesses a register in the processor part 102 or when the CPU 8 accesses
the local memory 3 or the frame memory 4 connected to the local bus 2.
[0031] 13 designates a clock generator for outputting a clock signal for adjusting the input/output
timing of the microprocessor 1 to the system bus 9 and for activating the CPU 8, the
system memory 10 and the bus arbitrator 11.
[0032] The address allocating method for the system, local and frame memories of FIG. 1
will now be described with reference to FIG. 2.
[0033] The two-port microprocessor divides a selfaddressable memory space into two halves
and then allocates them for a first system memory space 20 and a local memory space
40.
[0034] A second system memory space 30 is accessible only from the CPU.
[0035] 301 designates registers built in the two-port microprocessor. The registers 301
are allocated to addresses in the memory space of the CPU and is accessible from the
CPU. In this registers 301, an R0 register and an R1 register are used for dedicated
purpose.
[0036] Specifically, in FIG. 3, the R0 register is called a control register and stores
information about the operation control of the two-port microprocessor. Namely, the
most significant bit of the R0 register is called an operation control bit; if "0"
is stored, the two-port microprocessor is in an inoperative state, and if "1" is stored,
it assumes an operative state, in which microinstructions are being executed. 0 bit
to 23 bit is a field in which a work list address may be stored; the work list is
a list on which memory addresses of a drawing program and a graphics parameter to
be executed by the two-port microprocessor are described. When the two-port microprocessor
is to read the work list, the R0 register is used as an address register. But the
field which is effective as an address is only the range of 0 bit to 23 bits. The
R0 register is readable and writable as accessed from the CPU or as the instructions
are executed by the two-port microprocessor.
[0037] The R1 register is called a program counter, and stores the memory address of a microinstruction
to be read by the two-port microprocessor and increases the value automatically upon
reading of the instruction. In the R1 register, 0 bit to 23 bits are effective so
that the entire memory space of the two-port micro-processor may be used as a program
area.
[0038] The graphics drawing method of the apparatus of FIG. 1 will now be described.
[0039] In this apparatus, the CPU creates a work list and graphics parameters in a memory,
and the two-port microprocessor performs graphics drawing or the like with reference
to this work list and the graphics parameters.
[0040] FIG. 4 shows a memory map of commands and programs which are both needed for the
two-port microprocessor to perform a graphics drawing.
[0041] A work list 10-1 is a command stream or train which is to be given to the two-port
microprocessor from the CPU and which has leading addresses of the memory storing
the successive graphics parameters, and also leading addresses of the memory storing
the corresponding drawing programs, covering the entire graphics to be drawn.
[0042] 10-1-1 designates a leading address of the memory in which the parameters of a graphics
"c" are stored, while 10-1-2 designates a leading address of the memory in which the
drawing program of the graphics "c" is stored. In the addresses following the leading
address 10-1-2, the parameters and program of another graphics are stored. 10-1-3
designates a leading address of the parameters of a final graphics "h"; and 10-1-4,
a leading address of the drawing program of the final graphics "h". 10-1-5 designates
a final work list, so called "a list end". The value of the list end may be determined
optionally.
[0043] 10-2 designates the parameters of the graphics "c"; the number and sequence of the
parameters correspond to the drawing programs of the graphics "c".
[0044] Likewise, 10-3 designates the parameters of the graphics "h".
[0045] In the illustrated embodiment, the work list 10-1 and the graphics parameters 10-2,
10-3 are defined in the system memory 10; alternatively they may be defined in the
local memory 3.
[0046] 3-1 designates a work list reference program for reading a leading address of graphics
parameters from the work list to initiate the execution of graphics drawing programs
3-2-1, 3-2-2, 3-2-3.
[0047] In this embodiment, the programs 3-1, 3-2-1, 3-2-2, 3-2-3 are defined in the local
memory; alternatively they may be defined in the system memory 10.
[0048] The two-port microprocessor 1 stores an address of the work list in the R0 register
1101-1, and is usable as an address register when making a reference to the work list.
The R1 register 1101-2 stores a program address. The R2 register 1101-3 is usable
as an address register when reading the graphics parameterns.
[0049] The graphics drawing method of the two-port microprocessor will now be described
with reference to FIG. 5.
[0050] 50 designates a flowchart showing the procedure for the CPU to cause the two-port
microprocessor to start graphics drawing processing.
[0051] Firstly, the CPU creates a work list and graphics parameters in the memory (50-1).
The CPU then writes a leading address of the work list reference program in the R1
register (program counter) of the two-port microprocessor (50-2). Thereafter, the
CPU writes "1" in the operation control bit in the R0 register (control register)
of the two-port microprocessor, and also writes a leading address of the work list
in the work list address field of the R0 register (50-3). The work by the CPU has
thus been completed (50-4). Since the graphics drawing processing after this is assigned
to the two-port microprocessor, the CPU can perform another processing during that
time.
[0052] The two-port microprocessor starts its operation when the operation control bit of
the R0 register becomes "1".
[0053] In FIG. 5, reference numeral 60 designates a flowchart showing the work of the two-port
microprocessor. The two-port microprocessor executes the work list reference program
as the R1 register is set at the step 50-2. Firstly, the two-port micro-processor
reads data, with the work list address field of the R0 register being regarded as
a memory address, and writes the data in the R2 register (60-1). In addition, the
two-port microprocessor adds "1" to the value of the work list address of the R0 register.
Then the two-port microprocessor compares the content of the R2 register and the list
end value meaning the end of the work list (60-2). If they are equal, the two-port
microprocessor rewrites the operation control bit of the R0 register to "0" (60-3-b)
and thus assumes an inoperative condition (60-6). Otherwise, if they are not equal,
the two-port microprocessor reads data, with the work list address field of the R0
register being regarded as a memory address, and writes the data in the R1 register
(60-3-a). At the same time, "1" is added to the value of the work list address of
the R0 register. According to the value of the R1 register rewritten at the step 60-3-a,
the process of the two-port microprocessor is assigned to any of the graphics drawing
programs of the steps 60-4-a, 60-4-b, 60-4-c. With each of these programs, it is possible
to read the parameters from a memory with the R2 register being used as an address
register. Upon completion of the drawing program, the R1 register is rewritten to
a leading address of the work list reference program (60-5), returning to the process
of step 60-1. The two-port microprocessor executes the above processing until the
list end is detected from the work list.
[0054] The interior of two-port microprocessor will now be described with reference to FIG.
6.
[0055] 1101 designates an instruction fetch unit for fetching a microinstruction. The instruction
fetch unit 1101 has a built-in cache memory, and fetches the instruction from an external
memory if there exists no instruction corresponding to the cache memory.
[0056] 1102 designates a decoder for decoding a instruction code given from the instruction
fetch unit 1101 to control arithmetic operation.
[0057] 1103 designates an arithmetic or operation unit which is composed of registers and
an operation part and which executes an arithmetic and logic operation.
[0058] 1104 designates a memory access managing unit for processing a memory access request
generated in the two-port microprocessor and also an interruption of the CPU 8.
[0059] 102 designates a port 1 through which an access response from the CPU 8 and an access
to the system memory 10 are performed.
[0060] designates a port 2 through which accesses to the local memory 3 and the frame memory
4 are performed.
[0061] 1105 designates a bus switch for performing a selective bus connection between the
operation unit 1103 and the port 1 (102) and between the operation unit 1103 and the
port 2 (103).
[0062] Following is a detailed description of the individual units.
[0063] The instruction fetch unit 1101 is composed of a cache control part 1101-1, an instruction
fetch control part 1101-2, and an instruction buffer 1101-3. The pivot of the operation
control of the instruction fetch unit 1101 is the instruction fetch control part 1101-2,
which makes a decision on whether an instruction fetch should be executed or not,
depending on the state of the operation control bit of the R0 register 1103-1 in the
operation unit 1103. If the operation control bit is "0", the instruction fetch will
not be executed; if the instruction fetch is "1", the instruction fetch will be executed.
[0064] The instruction fetch control part 1101-2 issues a fetch request to the cache control
part 1101-1. The cache control part 1101-1 retrieves a cache memory in the cache control
part 1101-1, with the R1 register 1103-2 in the arithmetic unit 1103 serving as a
program address. If there exists an instruction in the cache memory, the cache control
part 1101-1 returns a response signal to the instruction fetch control part 1101-2,
along with the instruction code. The instruction fetch control part 1101-2 stores
the instruction in the instruction buffer 1101-3 and makes an increment of the program
address of the R1 register 1103-2 in the operation unit 1103. If there exists no instruction
in the cache memory, the cache control part 1101-1 issues an instruction fetch request
to the memory access managing unit 1104.
[0065] The decoder 1102 decodes an instruction given from the instruction fetch unit 1101,
controls the reading and writing of the registers in the operation unit 1103, and
also controls the operations therein. Further, in the case of a load or store (hereinafter
called "load/store") instruction, the decoder 1102 gives an load/store signal and
a port number to the memory access managing unit 1104. In the case of a load instruction,
it gives to the memory access managing unit 1104 the number (load register number)
of a register in which loaded data are to be written. The procedure discussed here
is represented by reference character A in FIG. 6. The port number is a signal for
designating whether a memory to access should be on the port 1 (102) or the port 2
(103).
[0066] The port designating method will now be described with reference to FIGS. 7 and 8.
[0067] FIG. 7 shows a port number designating circuit 1102-1 in the decoder 1102. As the
port number designating circuit 1102-1 decodes a load/store instruction, a selection
circuit 1102-1-a selects a value of one register designated from n number of registers
in the operation unit 1103 by an address register number, and inputs the selected
register value to a comparator 1102-1-b. To another input of the comparator 1102-1-d,
comparison data 1102-1-c are inputted to be compared with the value according to a
condition bit 1102-1-b, and the result of the comparison is outputted as a port number.
The comparison data 1102-1-c correspond to an border address between the system memory
space 20 and the local memory space 40 of the two-port microprocessor of FIG. 2, while
the condition bit 1102-1-b makes a decision as to which side of the border address
the system memory should be allocated to.
[0068] The relation between the condition bit 1102-1-b and the port number designating method
is shown in a table of FIG. 8. If the condition bit is "0", the port 1 is designated
when the data of the address register are equal to or larger than the comparison data,
and otherwise the port 2 is designated. Reversely, if the condition bit is "1", the
port 2 is designated when the data of the address register are equal to or larger
than the comparison data, and otherwise the port 1 is designated. Either the comparison
data or the condition bit is optionally settable so that the memory map of the memory
space accessible by the two-port microprocessor can be optionally determined.
[0069] Returning to FIG. 6, when the decoder 1102 requests the memory access managing unit
1104 for load/store, a wait signal (D) is returned if its port is in use or reserved.
Thus the instruction fetch unit 1101 and the decoder 1102 enter a wait status. As
the load/store request is accepted, the decoder 1102 decodes a succeeding instruction.
[0070] In the case of a load instruction, as data are read from the memory, the memory access
managing unit 1104 gives the decoder 1102 a wait signal (D), a register forced write
signal and its register number. At that time the decoder 1102 holds the execution
of an instruction in decodes so that the data read from the memory are written in
the register with priority over the others. This circuit is shown in FIG. 9.
[0071] In FIG. 9, the reading and writing of the register by instructions are executed by
an instruction decoding circuit 1102-2 which gives register numbers to control circuits
1102-3, 1102-4, 1102-5, 1102-6, 1102-7. Upon receipt of a register forced access signal
and a wait signal from the memory access managing unit 1104, the instruction decoding
circuit 1102-2 stops executing the instruction, and a register read control circuit
4 (1102-6) or a resister write control circuit 1102-7 executes reading from or writing
in a register designated by the forced access register number.
[0072] Until data are written from a memory to a register after a load instruction is accepted
by the memory access managing unit, there is the possibility that the register is
used according to the execution of a subsequent instruction. In this case, the execution
of such instruction must be waited until data are loaded from the memory into the
register. To this end, the decoder 1102 gives a read register number to the memory
access managing unit 1104 earlier than executing the instruction, for comparison with
a load register number. If the two values agree with each other, a wait signal (D)
is returned to wait executing the instruction.
[0073] The operation unit 1103 will now be described with reference to FIG. 6 again.
[0074] As discussed above, the RO register 1103-1 and the R1 register 1103-2 among n number
of registers are for dedicated use, and the R2 register to Rn-1 register are general
registers 1103-3. A barrel shifter 1103-4 combines two sets of 32-bit data to provide
64-bit data, and selects 32-bit data therefrom while shifting the 64-bit data a number
of bits which is designated by the 6-bit shift number. A three-operand logic unit
(three-operand LU) 1103-5 performs 256 kinds of logical operations between three input
data. An arithmetic and logic unit (ALU) 1103-6 performs arithmetic and logic operations
between two input data. Internal buses include four buses (R3B, R2B, R1B, R0B) for
reading and a single bus (WB) for writing; among these internal buses, the bus R1B
is used to provide the shift number of the barrel shifter 1103-4 and to designate
the operation mode of the three-operand LU 1103-5. When executing a store instruction,
the bus R2B is used for address transfer, and the bus R3B is used for data transfer.
When executing a load instruction, the bus R2B is used for address transfer, and the
bus WB is used for data transfer.
[0075] Now the memory access managing unit 1104 will be described with reference to FIG.
10.
[0076] In connection with a load/store managing circuit 1104-2, 1104-2-a means a 3-bit status
register 1 which indicates the state of execution with the port 1; F1, an instruction
fetch; L1, load; and S1, store. If "1" is set at any bit in the register 1, it means
that the port 1 is in operation or working associated with the bit of "1". 1104-2-b
designates a register for storing the number of a register in which data are to be
written by executing a load instruction.
[0077] Likewise, for the port 2, the load/store managing circuit 1104-2 includes a status
register 2 (1104-2-c) and a load register number register 2 (1104-2-d). But the status
register 2 (1104-2-c) is a 4-bit register and has, in addition to bits F2, L2, s2,
a bit I that means under execution of interface.
[0078] When a load/store request and a port number are given from the decoder 1102, the
load/store managing circuit 1104-2 checks the status registers 1104-2-a, 1104-2-c
for the designated port, and if the port is in operation, the managing circuit 1104-2
immediately gives to an OR circuit 1104-5 a signal indicating that the port is in
use. As a result, a wait signal is outputted to wait executing a load/store instruction
that is about to be executed. In every bit of the status registers 1104-2-a, 1104-2-c
is "0", the bit of a load or a store is set to "1". Further, in the case of a load
request, a load register number outputted from decoder is written in the load register
number registers 1104-2-b, 1104-2-d of the designated port.
[0079] The load/store managing circuit 1104-2 requests a load or store operation to the
designated port, and outputs a control signal for writing an address in an address
buffer in the port. In the case of store, it also outputs a control signal for writing
data in a data buffer in the port.
[0080] A memory access occurs not only in the case of a load/store instruction, but when
a mishit of an instruction cache is happened. The instruction fetch unit outputs an
instruction fetch request and a port number. The load/store managing circuit 1104-2
checks the status registers 1104-2-a, 1104-2-c of the designated port. If the port
is not in use, the load/store managing circuit 1104-2 gives to the instruction fetch
unit a signal to read a program address, and writes the program address in an address
buffer in the port.
[0081] When an acknowledge is returned from the designated port, the load/store managing
circuit 1104-2 clears or resets the store bit S1 or S2 of the status registers 1104-2-a,
1104-2-c of the port to "0" in the case of a store instruction. In the case of a load
instruction, the load/store managing circuit 1104-2 instructs the register forced
access circuit 1104-3 to perform such processing, and clears the load bit L1 or L2
of the status registers 1104-2-a, 1104-2-c of the designated port to "0".
[0082] The load/store managing circuit 1104-2 is also operable at the time of local bus
access by the CPU. Upon receipt of a bus interface request from the port 1, the port
2 will be unused, whereupon the managing circuit 1104-2 sets the bit I of the status
register 2 (1104-2-c) and returns a bus interface ready to the port 1. When the bus
interface request is canceled upon termination of using the local bus by the CPU,
the managing circuit 1104-2 clears the interface bit I of the status register 1104-2-c.
[0083] 1104-3 designates a register forced access circuit. After data have been written
from the memory to the data buffer in the port by the execution of the load instruction,
the register forced access circuit 1104-3 accepts a load register request and a load
register number from the load/store managing circuit 1104-2 and causes the decoder
1102 to execute a forced register writing. At that time the register forced access
circuit 1104-3 also controls the reading of the data buffer in the port.
[0084] The register forced access circuit 1104-3 is also operable when the CPU has accesses
to the register in the two-port microprocessor. Upon accept of the register write
or read signal and the register number from the CPU, the register forced access circuit
1104-3 causes the decoder 1102 to execute a forced register writing or reading. At
that time the register forced access circuit also controls the writing and reading
of a data buffer 1202-1 which temporarily stores reading data to or writing data from
the CPU in the port 1.
[0085] 1104-4 designates a bus switch control circuit for performing a bus connection control
for data transfer between the operation unit and each of the ports during the load/store
executing.
[0086] 1104-1 designates a register read inhibition circuit for accepting from the decoder
1102, before execution of the succeeding instruction, the register number to be read,
and compares the register number with the load register number held by the load/store
managing circuit 1104-2. Because there are four read buses in the operation unit 1103,
the register read inhibition circuit 1104-4 compares at most four read register numbers
and the load resister number. If there is an agreement in number, the register read
inhibition circuit 1104-4 prohibits executing the instruction that is about to be
executed.
[0087] The port 1 (102) will now be described with reference to FIG. 6 again.
[0088] The port 1 (102) is composed of a memory access unit 1201, a CPU response unit 1202,
an address latch 1203, a data latch 1204, and terminals 1205.
[0089] Upon receipt of a load/store instruction (G-1) from the memory access managing unit
1104, the memory access unit 1201 executes accessing to the system memory 10. In order
to hold the address and store data given from the operation unit 1103, the memory
access unit 1201 has an address buffer 1 (1201-1) and a W data buffer 1 (1201-2).
The memory access unit 1201 has also an R data buffer 1 (1201-3) for holding the data
read from the system memory 10.
[0090] The memory access unit 1201 acquires the right of use of the system bus 9 before
accessing a memory. Upon receipt of a response from the system memory 10 after having
started accessing with the right of use of the system bus 9, the memory access unit
1201 returns an acknowledge (G-2) to the memory access managing unit 1104. In the
case of load, the R data buffer 1 (1201-3) is read (G-3) by the memory access managing
unit 1104.
[0091] The CPU response unit 1202 is operable when the CPU 8 accesses the registers 1103-1
to 1103-3 in the two-port microprocessor and also when the CPU 8 accesses the local
memory 3 or the frame memory 4.
[0092] In the case of access to the registers, the CPU response unit 1202 outputs a register
access signal and a register number (H-1) to the memory access managing unit 1104.
In the case of writing in the registers, the memory access managing unit 1104 causes
the instruction fetch unit 1101 and the decoder 1102 to wait operating, and reads
(H-2) the data of the CPU data buffer 1202-1 to the bus MRB to perform forced writing
to the designated register. In the case of reading the register, the memory access
managing unit 1104 causes also the instruction fetch unit 1101 and the decoder 1102
to wait operating, and forcedly reads the register to perform writing (H-3) to the
CPU data buffer 1202-1. The CPU response unit 1202 writes the data in a latch 1204
and returns an acknowledge to the CPU, thus causing the CPU to perform data reading.
[0093] When the CPU accesses the local memory 3 or the frame memory 4, the CPU response
unit 1202 outputs a bus interface request (I-1) to the memory access managing unit
1104. When the local bus 2 will be free in the next memory cycle, the memory access
managing unit 1104 returns a bus interface ready (I-2) to the CPU response unit. The
CPU response unit 1202 outputs a load or store signal to a memory access unit 1301
and transfers an address from a latch 1203 in the port 1 (102) to a latch 1302 in
the port 2 (103). In the case of store, the CPU response unit 1202 transfers data
from the latch 1204 in the port 1 (102) to a latch 1303 in the port 2 (103). Accessing
to the local memory 3 or the frame memory 4 is executed by the memory access unit
2 (1301). In the case of store, when the memory accessing is started, the CPU response
unit 1202 returns an acknowledge to the CPU 8 to terminate the bus cycle of the CPU
8. In the case of load, after completion of the memory accessing, the CPU response
unit 1202 transfers to the latch 1204 in the port 1 (102) the data read in the latch
1303, and returns an acknowledge to the CPU 8, thus causing the CPU 8 to read the
data from the system bus 9.
[0094] Now, the port 2 (103) will be described.
[0095] The port 2 (103) is composed of a memory access unit 2 (1301), an address latch 1302,
a data latch 1303, and terminals 1304.
[0096] Upon receipt of requests from the memory access managing unit 1104 or the CPU response
unit 1202, the memory access unit 2 (1301) executes accessing to the local memory
3 or the frame memory 4. In order to hold the address and store data given from the
operation;unit 1103, the memory access unit 2 (1301) has an address buffer 2 (1301-1)
and a W data buffer 2 (1301-2). Further, in order to hold the data read from the local
memory 3 or the frame memory 4, the memory access unit 2 (1301) has also an R data
buffer 2 (1301-3). The local bus 2 is a synchronous bus which uses a clock of sextuple
the internal clock for a single memory cycle.
[0097] The simultaneous access to two memories of the two-port microprocessor will now be
described with reference to FIG. 11.
[0098] Assume that an instruction n fetched from an instruction chache with an internal
clock k is a load instruction, with the port 1 designated, as the result that the
instruction is decoded. The memory access managing unit checks the operation state
of the port 1 and, if discriminated that the port 1 is free, outputs a load 1 (designated
by 1 in FIG. 11) to the memory unit 1. Further, the operation unit reads onto the
bus the content of the register designated by the address register, and the memory
access managing unit connects the bus switch to write (2) in the address buffer 1
an address on the bus.
[0099] The memory access unit 1 makes a request for acquiring the system bus, causes the
address buffer 1 to output the memory address to the system bus, when a right to use
the bus is given as indicated by an arrow 3, to execute a reading access to the system
memory. When a response is returned from the system memory, the memory access unit
1 returns an acknowledge 1 (4) to the memory access managing unit. The memory access
unit 1 also reads data from the system bus to an R data buffer 1, as indicated by
an arrow 5. Upon receipt of the acknowledge 1 (4), the memory access managing unit
performs a forced writing over the register in the operation unit to cancel (6) the
load 1.
[0100] Also assume that an instruction n + 1 fetched with an internal clock k + 1 is a load
instruction. The memory access managing unit checks the operation state of the port
2, and outputs a load 2 (11) to the memory access unit 2. Further, the operation unit
reads in the bus the content of the register designated by the address register, and
the memory access managing unit writes (12) it in the address buffer of the memory
access unit 2. The memory access unit 2 outputs (13) the memory address in the memory
cycle starting at a clock k + 7, and reads (15) the data in the R data buffer 2 at
a clock k + 11. The memory access managing unit reads the R data buffer 2 in response
to an acknowledge 2 (14) of clock k + 10, and performs forced writing on the register
in the operation unit.
[0101] Accordingly, in the two-port microprocessor, since the memory access managing unit
manages the operation states of the two ports and has the function of performing register
forced writing of the data read from the memory, memory accesses simultaneously using
two ports can be achieved.
[0102] The register access in the two-port microprocessor by the CPU will now be described.
[0103] FIG. 12 is a timing diagram showing the manner in which the CPU reads the content
of the R0 register in the two-port microprocessor, modifies it and writes it back
in the R0 register. When an external decoder 12 receives a register read request from
the CPU, the CPU response unit outputs a CPU register read signal (1 in FIG. 12) and
a register number to the memory access managing unit. Immediately in response to this,
the memory access managing unit causes the instruction fetch unit and the decoder
to wait operating (an instruction j + 1 is caused to wait in FIG. 12), forcedly reads
the R0 register in the operation unit, and writes (2) the data in the CPU data buffer.
The CPU response unit outputs the data of the CPU data buffer to the system bus and
returns a response to the CPU. The CPU modifies the read data and subsequently starts
a writing cycle.
[0104] Upon receipt of a register write request from the CPU, the CPU response unit reads
(11) the data on the system bus into the CPU data buffer, and outputs (12) a CPU register
write signal and a register number to the memory access managing unit. The memory
access managing unit immediately causes the instruction fetch unit and the decoder
to wait operating (an instruction k + 1 is caused to wait in FIG. 12), and performs
a forced writing (13) in the R0 register in the operation unit.
[0105] As is apparent from the execution stages of a microinstruction shown in FIG. 12,
forced reading (2) is executed in a cycle subsequent to the cycle in which the CPU
register read signal (1) is given, and forced reading (13) is executed in a cycle
next to the cycle subsequent to the cycle in which the CPU register write signal (12)
is given.
[0106] The accessing to the local memory and the frame memory by the CPU will now be described.
[0107] FIG. 13 is a timing diagram showing the manner in which the CPU makes a read access
to the local memory. When a local memory access request from the CPU is received by
the external decoder, the CPU response unit outputs a bus interface request (1) to
the memory access managing unit.
[0108] The memory access managing unit checks the operation state of the port 2. As an acknowledge
2 (2) is returned from the memory access unit 2 at a clock k + 4 and as the status
register of the port 2 in the memory access managing unit is cleared, the memory access
managing unit receives a bus interface request, sets the bit I of the status register
and returns (3) a bus interface ready to the CPU response unit. The CPU response unit
cancels (4) the bus interface request, transfers to an address latch in the port 2
the memory address read in an address latch in the port 1, and issues a load request
to the memory access unit 2. The memory access unit 2 reads the data from the local
memory and returns (6) an acknowledge to the CPU response unit. The CPU response unit
transfers to a data latch in the port 1 the data read in a data latch in the port
2 and outputs (7) it onto the system bus. Upon receipt of the response from the port
1, the CPU reads the data on the system bus.
[0109] The read/modify/write instruction of the two-port microprocessor will now be described.
[0110] This instruction is an instruction for performing an operation over the data read
from the memory and for writing the result of the operation back to the original memory
address at two successive memory cycles. FIG. 14 shows the signal transfer between
the units during the read cycle of the read/modify/write instruction. FIG. 15 shows
the signal transfer between the units during the subsequent modify/write cycle. FIG.
16 is a timing diagram showing the operations of the various parts of FIGS. 14 and
15. FIG. 17 shows the concept of the microprocessor of this invention using a modify
buffer.
[0111] First of all, the read cycle will be described with reference to FIGS. 14 and 16.
[0112] Having decoded the read/modify/write instruction (1 in FIGS. 14 and 16) fetched with
the internal clock k, the instruction decoder 1102 immediately returns a wait signal
(2) to the instruction fetch unit. This signal continues being outputted to the operation
unit until the modify instruction is executed. The instruction decoder causes the
instruction fetch unit to output the content of Ri register as the memory address
onto the bus R2B, and gives (3) a load request, a store request and a port number
to the memory access managing unit. Further, the instruction decoder causes the operation
unit to output the content of Rs register onto the bus R3B to fetch (4) the data into
the modify buffer 1102-9 in the instruction decoder. Meanwhile, upon receipt of a
load/store request, the memory access managing unit sets to "1" both a load bit and
a store bit of the status register in the designated port, outputs a load signal (7),
controls (5) the bus switch to write (6) the memory address in the address buffer.
The designated port latches (8) the memory address in the address buffer, outputs
(9) it to the memory, and starts the read cycle. When data are read from the memory,
the port latches (10) it, transfers (11) it to the R data buffer, and returns an acknowledge
(12) to the memory access managing unit. The memory access managing unit gives (13)
a forced write signal to the instruction decoder, controls the bus switch to write
(14) in Rq register the data read from the R data buffer.
[0113] The subsequent processes will now be described with reference to FIGS. 15 and 16.
[0114] The memory access managing unit resets the load bit of the status register to "0",
and gives (15 in FIGS. 15 and 16) a store signal to the port. The port latches (16)
a memory address in the address buffer and outputs (17) it to the memory to start
the write cycle.
[0115] The instruction decoder cancels (18) the wait signal to the instruction fetch unit,
and causes the operation unit to execute (19) the modify/buffer instruction. This
example presents an instruction for performing an operation in the ALU or the contents
of the Rq register and the Rr register and for storing the result of the operation
in the Rq register.
[0116] The memory access managing unit controls (20) the bus switch at the timing at which
the ALU outputs the result of the operation, so as to write (21) the data on the bus
WB into the W data buffer in the port.
[0117] The port latches (22) the data in the W data buffer and outputs (23) the data to
the memory. Further,! the port returns an acknowledge (24) to the memory access managing
unit and clears the store bit of the status register to "0".
[0118] As shown in the block diagram of FIG. 17, independently of the ordinary instruction
buffer for holding the instruction of a program, a modify buffer may be provided for
holding a sub-instruction preset in any one of the registers. If the execution of
the sub-instruction is needed as a result of decoding an instruction in the instruction
buffer, an instruction in the modify buffer may be immediately decoded corresponding
to the result of the instruction decoding of the instruction buffer, so that the result
of modification of the data read in the read cycle can be in time for the next write
cycle, as is apparent from the timing diagram of FIG. 16. Further, the instruction
to be fetched in the modify buffer is readable from an optional one of the registers
R2 through Rn-1. In FIG. 17, the modify buffer is illustrated as an independent block
from the block of the decoder for clarity of the concept.
[0119] The rectangular transfer method from the system memory to the frame memory will now
be described with reference to FIG. 18.
[0120] 10-1 represents the system memory 10 in an X-Y coordinate space. A source memory
width MWS means the width of the X-Y coordinate space 10-1 of the system memory that
is a source of the transfer. ORGS represents the origin of the X-Y coordinate space
10-1, and its value is a bit address on the system memory 10. At the source rectangular
field 10-2, Xss and Yss represent X coordinate and Y coordinate, respectively, of
the transfer start position; and Xse and Yse represent X coordinate and Y coordinate,
respectively, of the transfer end position.
[0121] 4-1 represents the frame memory 4 in the form of an X-Y coordinate space. A destination
memory width MWD means the width of the X-Y coordinate space 4-1 of the frame memory
that is a destination of the transfer. ORGD represents the origin of the X-Y coordinate
space 4-1, and its value is a bit address on the frame memory 4. At the destination
rectangular field 4-2, Xds and Yds represent X coordinate and Y coordinate, respectively,
of the transfer start position.
[0122] Firstly, as indicated by 50 in FIG. 18, the CPU 8 generates in the system memory
10 a work list composed of a leading address of a graphics transfer program and a
leading address of graphics parameters. The CPU 8 also generates parameters needed
to graphics transfer at successive addresses following to the pa-rameters' leading
address in the work list.
[0123] Then, having been activated (60) by the CPU 8, the two-port microprocessor 1 reads
from the work list the leading address of a program to be executed and thereby fetches
(70) from the local memory 3 a graphics transfer program. The source address of rectangular
field is calculated from the coordinate values read from the system memory 10, ORGS
and the source memory width (80). Likewise, the destination address of rectangular
field is calculated from the coordinate values read from the frame memory 4, ORGD
and the destination transfer memory width (80).
[0124] In order to perform a data transfer, with a logical operation, between data of a
source rectangular field and data of a destination rectangular field, a data operation
is executed, by utilizing the read/modify/write instruction discussed in connection
with FIGS. 14 through 16, between the source data (90-1) read from the system memory
10 and the destination data (90-2) in the frame memory 4. The result of this logical
operation is written (90-3) in the destination address. By executing the address calculation
and the data operation over the rectangular fields between which the data are to be
transferred, a high-speed transfer of the rectangular field data from the system memory
10 to the frame memory 4 can be achieved.
[0125] The general construction of the operation unit is discussed above in connection with
FIG. 6. FIG. 19 shows the detailed construction of the operation unit to be used for
this data transfer of rectangular field. The operation unit comprises a register for
storing the number of words (memory width) in X direction of a bit-map memory, registers
(a source address register, a pattern address register, and a destination address
register) for storing word addresses of three rectangular fields (a source field,
a pattern field, and a destination field) defined on the bit map memory, a source
shift number register for storing a difference in bit position between the source
field and the destination field, a pattern shift number register for storing a difference
in bit position between the pattern field and the destination field, a barrel shifter
for shifting source data based on the source shift number register or shifting pattern
data based on the pattern shift number register, a three-operand logic unit (three-operand
LU) for performing a logical operation of the shifted source data, pattern data and
destination data, a mask means for designating a writing inhibition field at opposite
ends of the destination field, and a composer circuit for composing the operational
result and the destination data based on the mask means.
[0126] According to this invention, partly since two sets of buses of the two-port microprocessor
are usable for instruction fetch and data access, and partly since continuous memory
accessing is possible, the memory accessing efficiency is improved. In the case of
a graphics processing apparatus in particular, it is possible to transfer the graphics
data between the system memory and the frame memory at high speed. Further, a read/modify/write
process frequently used in graphics processing can be executed with no free or empty
cycle between the read cycle and the write cycle, improving the performance of a graphics
processing apparatus.
[0127] Although the preferred embodiments of this invention have been described here, it
should be understood that various other modifications to and alternative arrangements
of the elements of the illustrated embodiments may be suggested.
[0128] For example, the microprocessor using a modify buffer should by no means be limited
to the two-port processor, but may be a conventional one-port processor. Further,
although the graphics processing apparatus is illustrated here in the foregoing embodiments,
this invention may be used in a dedicated purpose processor for special processes,
e.g. printer control or communication control, which should be accompanied by data
transfer between separate buses.